Unsupervised adaptation of an acoustic model using confidence measures based on phoneme posterior probabilities

摘要

In this paper, we study on an accurate unsupervised adaptation method for spontaneous speech recognition. In unsupervised adaptation framework, the effectiveness of adaptation process is greatly affected by the mis-recognized labels. Therefore, selection of the adaptation data guided by the confidence measures is effective in unsupervised adaptation. We propose an phoneme error minimization framework for accurate phoneme-labels and use of phoneme-level confidence measures for improved unsupervised adaptation. Experimental results showed that the proposed method could reduce the mis-recognized labels in the adaptation process, and consequently improved the adaptation accuracy. Furthermore the selection of the adaptation data using the phoneme confidence measures improved the adaptation accuracy.